Image enhancement network model pruning method and device, electronic equipment and storage medium
By evaluating candidate pruning structures using a performance detector and analyzer model in the pruning search space of the image augmentation network model, the problem of simplifying the model structure in the prior art is solved. This achieves a reduction in computational and storage overhead while maintaining image augmentation capabilities, thereby improving the pruning efficiency of the image augmentation network.
CN122198015APending Publication Date: 2026-06-12NILIAN RUITAI INFORMATION TECHNOLOGY (SHANGHAI) CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NILIAN RUITAI INFORMATION TECHNOLOGY (SHANGHAI) CO LTD
- Filing Date
- 2026-03-16
- Publication Date
- 2026-06-12
Smart Images
Figure CN122198015A_ABST
Abstract
The application discloses a kind of image enhancement network model pruning method, device, electronic equipment and storage medium, it is related to network model simplification field.The method comprises: at least two candidate pruning structures are determined from the pruning search space of image enhancement network model by performance detector model;For each candidate pruning structure, respectively extract basic structure features, structure expression ability features and network complexity features;According to basic structure features, structure expression ability features and network complexity features, determine the image enhancement score corresponding to the candidate pruning structure by performance analyzer model;According to image enhancement score, determine target pruning structure from candidate pruning structure;If target pruning structure satisfies preset termination condition, then according to target pruning structure, prune image enhancement network model, obtain pruned image enhancement network model.The above technical solution simplifies model structure while ensuring image enhancement capability, improves image processing efficiency.
Need to check novelty before this filing date? Find Prior Art